In this context, are bias and variance always related to MSE, or only when the cost function for the statistical learning method is related to MSE?

  • 2
    $\begingroup$ In which context? $\endgroup$ – kjetil b halvorsen Sep 2 '16 at 12:03
  • $\begingroup$ In the context of machine learning "bias variance tradeoff" question $\endgroup$ – user_anon Sep 2 '16 at 12:11
  • 1
    $\begingroup$ Although this question seems to be arising from a confusion, it is possible to answer (+1 to @StudentT below). IMO, it can stay open. $\endgroup$ – gung Sep 2 '16 at 12:28

I think you're totally confused. Bias/variance relates how well your model fits relative to overfitting and underfitting. MSE is simply one of the many possible measures we can use to quantify how well a model performs.

High bias can be measured by very high MSE and high variance can be measured by very low MSE.

  • 2
    $\begingroup$ Low bias would normally be associated with overfitting. Think about Lasso or ridge regression where bias is induced to avoid overfitting. $\endgroup$ – Richard Hardy Sep 2 '16 at 12:35
  • $\begingroup$ @RichardHardy I think I've given the wrong orders, edited. $\endgroup$ – SmallChess Sep 2 '16 at 12:45
  • $\begingroup$ I was looking on wikipedia at "Bias–variance decomposition of squared error" from here. I thought bias and variance "exist" only when we talk about MSE. So, reading your answer, it is not. Thanks! :) $\endgroup$ – user_anon Sep 2 '16 at 13:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.